CHAPTER2

Methods and Principles of Biological Systematics

iological systematics (or taxonomy) is the theory and practice of Bgrouping individuals into species, arranging those species into larg- er groups, and giving those groups names, thus producing a classification. Classifications are used to organize information about plants, and keys can be constructed to identify them. There are many ways in which one might construct a classification. For example, plants could be classified on the basis of their medicinal proper- ties (as they are in some systems of herbal medicine) or on the basis of their preferred habitat (as they may be in some ecological classifications). A phylogeny-based classification, such as that followed in this book, attempts to arrange organisms into groups on the basis of their evolution- ary relationships. There are two main steps in producing such a classifica- tion. The first is determining the phylogeny, or evolutionary history. The second is basing the classification on this history. These two steps can be, and often are, separated, such that every new theory of relationships does not lead automatically to a new classification. This chapter will outline how one goes about determining the history of a group, and then will dis- cuss briefly how one might construct a classification, given that history.

What Is a Phylogeny? As described in Chapter 1, evolution is not simply descent with modifica- tion, but also involves the process of separation of lineages. Imagine for a moment a population of organisms that all look similar to each other. By some process, the population divides into two populations, and these two populations go on to evolve independently. In other words, two lineages (ancestor–descendant sequences of populations) are established. We know this has happened because the members of the two new populations acquire, by the process of mutation, new characteristics in their genes, and possibly changes in their overall form, making the members of one popu- lation look more similar to each other than to members of the other popu- lation or to the ancestral population. These characteristics are the evidence for evolution.

9 10 CHAPTER TWO

For example, a set of plants Year 10 will produce offspring that are Year 9 genetically related to their par- Year 8 ents, as indicated by the lines in Figure 2.1. The offspring will Year 7 produce more offspring, so that Year 6 we can view the population over Year 5 several generations, with genetic Year 4 connections indicated by lines. Year 3 If the population divides into Year 2 two separate populations, each will have its own set of genetic Year 1 connections, and eventually will Petals white, stems herba- Petals white, stems woody, Petals red, stems herba- acquire distinctive characteris- ceous, leaves non-hairy, leaves non-hairy, stamens five, ceous, leaves non-hairy, tics. For example, the population stamens five, fruit dry, fruit dry, seed coat smooth stamens five, fruit dry, on the right could develop red seed coat smooth seed coat smooth flowers, whereas the stems of the population on the left could Figure 2.1 The evolution of two hypothetical species. Each circle represents a plant. A mutation in the lineage on the left causes a change to woody stems, which is then become woody. Red flowers and transmitted to descendant plants.Woody-stemmed plants gradually replace all the woodiness are evidence that each herbaceous ones in the population. A similar mutation in the lineage on the right leads of the two populations consti- to a group with red petals. tutes a single lineage. The same process can repeat, and each of the new populations can divide (Figure 2.2). Again, we states. In this case, the character “flower color” has two know this has happened because of a new set of charac- states, white and red. The character “stem structure” also teristics acquired by the newly formed populations. has two states, woody and herbaceous, and so forth. All Some of the woody plants have fleshy fruits, and anoth- else being equal, plants with the same state are more er group has a spiny seed coat. Meanwhile, some of the likely to be related than those with different states. red-flowered plants now have only four stamens, and The critical point in this example, however, is that another set of red-flowered plants have hairy leaves. characteristics such as red petals and woody stems are The characteristics of plants, such as flower color or new, and they are derived relative to the ancestral popu- stem structure, are generally referred to as characters. lation. Only such new characters tell us that a new lin- Each character can have different values, or character eage has been established; retaining the old characteristic

Year 18 Petals white, stems herbaceous, 17 leaves non-hairy, stamens five, fruit dry, seed coat smooth 16 15 Petals white, stems woody, 14 leaves non-hairy, stamens five, fruit dry, seed coat smooth 13 12 Petals white, stems woody, 11 leaves non-hairy, stamens five, fruit dry, seed coat spiny 10 9 Petals white, stems woody, leaves non-hairy, stamens five, 8 fruit fleshy, seed coat smooth 7 Petals red, stems herbaceous, 6 leaves non-hairy, stamens five, fruit dry, seed coat smooth 5 Petals red, stems herbaceous, 4 leaves hairy, stamens five, 3 fruit dry, seed coat smooth 2 Petals red, stems herbaceous, leaves non-hairy, stamens four, 1 fruit dry, seed coat smooth Figure 2.2 The same hypothetical set of plants as in Figure 2.1 after eight years and two more speciation events. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 11

(white flowers, herbaceous stems, non-hairy leaves, five groups, one with herbaceous stems and red petals, the stamens, dry fruit, smooth seed coat) does not tell us other with woody stems and white petals. Each of these anything about what has happened. groups can also be divided into two groups. Thus the A character state that is derived at one point in time classification can be derived directly from the phylogeny. will become ancestral later. In Figure 2.2, woody stems Note that the hierarchy is not changed by the order in are derived relative to the original population, but are which the branch tips are drawn. The shape, or topolo- ancestral relative to the groups with fleshy fruits or gy, of the tree is determined only by the connections spiny seed coats. between the branches. We can tell the evolutionary A group composed of an ancestor and all of its “story” by starting at any point in the tree and working descendants is known as a monophyletic group (mono, up or down. This means that the terms “higher” and single; phylum, lineage). We can recognize it because of “lower” are not really meaningful, but simply reflect the shared derived characters of the group (synapo- how we have chosen to draw the evolutionary tree. morphies). These are character states that have arisen From this point of view, a plant systematics course could in the ancestor of the group and are present in all of its as well begin by covering the Asteraceae, which some members (albeit sometimes in modified form). This textbooks consider an “advanced” family, and then concept was first formalized by Hennig (1966) and working out to other members of the asterid clade, Wagner (1980). instead of starting with the so-called “primitive” fami- The diagrams of Figures 2.1 and 2.2 are cumbersome lies, such as Magnoliaceae and Nymphaeaceae. The lat- to draw, but can be summarized as a branching tree (Fig- ter simply share a set of characters thought to be ances- ure 2.3A). It is also inconvenient to repeat the ancestral tral, but these are combined with a large set of derived character states retained in every group, so systematists characters as well. commonly note only the characters that have changed, with tick marks placed on the appropriate branches to indicate the relative order in which the character states Determining Evolutionary History originated (Figure 2.3B). CHARACTERS, CHARACTER STATES, AND The shared derived characters in Figure 2.3B can be NETWORKS arranged in a hierarchy from more inclusive (e.g. stems woody or petals red) to less inclusive (e.g., leaves hairy, In the example in Figures 2.1, 2.2, and 2.3, we have seed coat spiny). These then lead to the obvious conclu- described evolution as though we were there watching it sion that the plants themselves can be arranged in a hier- happen. This is rarely possible, of course, and so part of archical classification that is a reflection of their evolu- the challenge of systematics is to determine what went tionary history. The plants could be divided into two on in the past. The relatives of an extant species must be

(A) Petals red, stems herbaceous, Petals white, leaves non-hairy, Petals red, stems stems woody, stamens four, herbaceous, leaves non-hairy, fruit dry, seed Petals white, leaves hairy, stamens five, coat smooth stems woody, stamens five, fruit fleshy, leaves non-hairy, fruit dry, seed seed coat smooth stamens five, coat smooth fruit dry, seed Petals red, stems herbaceous, Petals white, coat spiny stems woody, leaves non-hairy, stamens five, leaves non-hairy, fruit dry, seed coat smooth stamens five, fruit dry, seed Petals white, stems herbaceous, coat smooth leaves non-hairy, stamens five, fruit dry, seed coat smooth

(B) Fruit fleshy Seed coat spiny Stamens four Leaves hairy

Stems woody Petals red Figure 2.3 (A) A simple way to redraw the pattern of change shown in Figure 2.2. Full descriptions are provided for each of Petals white, stems herbaceous, the ancestors and their descendants. (B) A simpler way to leaves non-hairy, stamens five, redraw Figure 2.3A, showing only the mutations that occurred fruit dry, seed coat smooth in the various lineages. 12 CHAPTER TWO determined by examining that species closely for charac- and it is impossible to summarize the literature here. teristics that are believed to be heritable. These can be Many phylogenetic systematists argue that homology any aspect of the plant that can be passed down geneti- can only be determined by constructing an evolutionary cally through evolutionary time and still be recogniz- tree, a viewpoint that will be followed in this text. When able. For example, petal color, inflorescence structure, reading the literature, it is worth checking what particu- and plant habit are all known to be under genetic con- lar authors mean when they use the term. trol. Although they often show considerable phenotypic From observing plants, groups that share particular variation, they are generally stably inherited from one characteristics can be identified. For example, a large generation to the next and thus would provide good tax- group of plant species has pollen with three grooves, or onomic characters. Many examples of such heritable germination furrows, called colpi; the pollen is thus characters are described in Chapter 4. Characters of described as tricolpate. Within this large group is a small- DNA and RNA can also be used, and are described in er group that has fused petals, and within the fused-petal Chapter 5. group is a still smaller group with flowers arranged in a It may be harder than you think to determine which head. These nested groups can be diagrammed as a set of structures in one plant can be compared to structures in ovals (a Venn diagram) as in Figure 2.4A, with the indi- another plant. Two structures may be similar because vidual shapes representing plant species. they are in a similar position in the different organisms, The same information can be drawn as a network or because they are similar in their cellular and histolog- (Figure 2.4B). In this case, number of colpi in pollen, ical structure, or because they are linked by intermedi- fusion of petals, and type of inflorescence are shown as ate forms (either intermediates at different developmen- vertical lines. All shapes (species) to the left of the pollen tal stages of the same organism or intermediates in line have fewer than three colpi, whereas everything to different organisms). These are Remane’s criteria of the right of the line has tricolpate pollen, as indicated by what he called homology, but what we are calling simi- the numeral 3. Likewise, the line for petals indicates a larity. The assessment of similarity is the basis of all of shift between free petals and fused petals, and the inflo- comparative biology and of systematics in particular. rescence line a shift between flowers clustered in a head Systematics entails the precise observation of versus flowers borne separately. organisms. Without accurate comparative morpholo- All the plants indicated by the same shape are gy, classification of any sort is impossible. Without drawn as though they arose at the same point in time. careful description of characters and their states, phy- This is because, for the purposes of this simplified logeny reconstruction and the description of history example, we have not provided any information on are meaningless. their order of origin. Determining similarity is the first step in determining We can determine the length of the network, which is homology, or identity by descent. Be aware, however, the number of changes. For example, proceeding from that the word homology has many different meanings, right to left, there is one change each in inflorescences,

(A) (C) Pollen Inflorescence grooves Petals in heads Pollen tricolpate Star plants < 3 free no Fused petals

Inflorescence Circle plants 3 free no a head Square plants 3 fused no

Diamond plants 3 fused yes

(B) Inflorescence Pollen colpi Petals a head < 3 3 free fused no yes

Figure 2.4 (A) Venn diagram of a set of plants. A large group has tricolpate pollen. Of those plants, a smaller group has fused petals, and of the plants with tricolpate pollen and fused petals, a subset has flowers arranged in a head. (B) The pattern of Figure 2.4A redrawn as an unrooted network. (C) The pattern of Figures 2.4A and B redrawn as a matrix. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 13 petals, and pollen grooves, so the network can be EVOLUTIONARY TREES AND ROOTING described as having a length of three. Figure 2.4 shows three different ways of recording and The same information can also be presented as a organizing observations about plants. Even though the matrix (Figure 2.4C). This time the rows correspond to network looks somewhat like a time line, it is not. It could plants, and the columns correspond to characters of the be read from left to right, right to left, or perhaps from the plants. The character states are then used to fill in the middle outward. To turn it into an evolutionary tree, we matrix. These are, or are hypothesized to be, genetic must determine which changes are relatively more recent changes that potentially distinguish the groups of plants and which occurred further in the past. In other words, the in the matrix. Thus the three changes in the network of tree must be rooted, which causes all character changes to Figure 2.4B represent three changes in character states or be polarized, or given direction. (Some workers distin- genes of plants. guish between an evolutionary tree, a phylogeny, and a In the example, we have implied that the determina- branching diagram or cladogram, but in this text the terms tion of the character states is perfectly obvious. This is are used interchangeably.) often not the case, however, particularly with morpho- If you imagine that the network is a piece of string, you logical characters. The variation among similar struc- can keep the connections exactly the same, even when tures must be described by dividing the character into you pull down a root in many different places. The net- character states. This is a hypothesis of underlying work from Figure 2.4B is redrawn in Figure 2.5, but rooted genetic control, although it is rarely framed this way. For in three different places. Notice that the length of each tree example, if two species differ in the color of their flow- (or cladogram) is the same as the length of the original ers, we may score the character petal color as having two network—three—and that all the connections are the states, red and blue. By scoring it this way, we are same, but that the order of events differs considerably. For hypothesizing that there are underlying genes that example, in the rooting shown in Figure 2.5A, the ances- switched, over evolutionary time, to produce red flow- tral plants had pollen with fewer that three colpi, petals ers from a blue-flowered ancestor, or blue flowers from a not fused, and flowers not in heads, whereas in Figure red-flowered ancestor. 2.5B, we would conclude that the ancestral plants had In this case, we know that there are genes (for example, exactly the opposite. In Figure 2.5C, the tree is rooted in components of the anthocyanin pathway) that do in fact such a way that the ancestor had tricolpate pollen. The control flower color, and thus the inference of two states pollen later changed to having fewer than three colpi in controlled by a genetic switch is probably a reasonable one lineage, whereas the other lineage kept the pollen guess. In most cases, however, we have no idea of the character state of three colpi and later acquired fused genetic control of the structural characters observed. In petals and flowers in heads. making hypotheses about the nature of the underlying As is obvious from inspection of Figure 2.5, the root- switches, then, often the only recourse is to be sure that ing of the tree is critical for interpreting how plants the character states really are distinct. For quantitative evolved. Different roots suggest different patterns of characters such as leaf length or corolla tube width, this changes (character polarizations). There has been much means drawing a graph to be sure that the species we are discussion among systematists of how the position of studying have measurements that do not overlap. For the root should be determined. One frequent suggestion many characters, the measurements do overlap, such that is that one should use fossils. But just because a plant has any guess as to the underlying switches, and therefore been fossilized does not mean that its lineage originated division into character states, is unsupported by any evi- earlier than plants now living; we know only that it died dence. In these cases, the characters should be omitted out earlier. In determining evolutionary history we are from the phylogenetic analysis (unless the overlap is interested in determining when lineages—taxa or taxo- caused by only a few plants, in which case the character nomic groups—diverged from one another (when taxa could be scored as polymorphic and retained in the analy- originated). When taxa die out is interesting to know, but sis). Even though such characters probably reflect genetic it is not relevant to determining origins. changes over evolutionary time, with our current knowl- In general, evolutionary trees are rooted using a rela- edge it is difficult to extract from them information on the tive of the group under study: an outgroup. When underlying changes, although methods of dealing with selecting an outgroup, one must assume only that the plants with variable characters have been developed. ingroup members (members of the group under study) Variability and overlap in morphological characters are more closely related to each other than to the out- are, of course, good reasons why many systematists group; in other words, the outgroup must have separat- have turned to molecular data in constructing phyloge- ed from the ingroup lineage before the ingroup diversi- nies. The recognition of molecular character states (i.e., fied. Often several outgroups are used. If an outgroup is nucleotides) is often easier and more precise, although added to the network, the point at which it attaches is even this can be difficult if gene sequences are hard to determined as the root of the tree. align, or if restriction fragments are similar in size (see In the case of the plants in Figures 2.4 and 2.5, the Chapter 5). plants shown are all flowering plants (angiosperms), 14 CHAPTER TWO

Figure 2.5 (A) One possible rooting (A) of the network in Figure 2.4B. Note that the number of evolutionary steps (character state changes) is the same yes as the unrooted network. (B) A second possible rooting of the same network. no Inflorescence

(C) A third possible rooting of the Time fused a head same network. free Petals 3

< 3 Pollen colpi

(B)

< 3

3 Pollen

Time colpi free Petals fused no

yes Inflorescence a head

(C)

yes < 3 no Inflorescence Time fused Pollen 3 a head colpi free Petals

and their closest living relatives are either the conifers, are monophyletic (i.e., they form a clade). In fact, the flow- cycads, gnetophytes, or ginkgos (see Chapter 7). In Fig- ering plants with fused petals and flowers arranged in a ure 2.6A, a conifer is added to the matrix from Figure head are the family Asteraceae, which are known to form 2.4C. (We could have used all “gymnosperms” as out- a monophyletic group. Thus having flowers in heads is a groups, but have chosen only one for simplicity of the synapomorphy for (is a shared derived character for, indi- example.) Because conifers do not have petals or flow- cates the monophyly of) the Asteraceae, having fused ers, two of the characters must be scored as not applica- petals is a shared derived character (synapomorphy) unit- ble, but we do know that conifer pollen does not have ing the square species with the diamond species, and hav- three colpi. With this information, the conifer can be ing tricolpate pollen indicates the monophyly of the circle added to the network as an outgroup, as in Figure 2.6B. plus square plus diamond species. Because it attaches among the star species, the tree can Notice how important rooting is for determining be rooted and redrawn as in Figure 2.6C. This corre- monophyly. If Figure 2.5B were the correct rooting of the sponds to the rooted tree in Figure 2.5A and strengthens flowering plant phylogeny, then fused petals and flow- the hypothesis that Figure 2.5A accurately reflects evolu- ers in heads would be ancestral character states (usually tionary history. called symplesiomorphies) rather than derived (syna- Note that the tree can be drawn in different ways and pomorphies). In this case, the species indicated by dia- still reflect the same evolutionary history. Comparing monds and squares would not share any derived charac- Figures 2.7A and B with Figure 2.6C shows that the ter. Notice also that a group including the diamond plus branches of the tree can be “rotated” around any one of square plants in Figure 2.5B does not include all the the branch points (nodes) and not affect the inferred descendants of their common ancestor—some of those order of events. descendants went on to become the circle and the star With a rooted tree (and only with a rooted tree), we can plants. Therefore, if Figure 2.5B were correct, then the determine which groups are monophyletic. Therefore, in diamond plus square species would not be a mono- the example laid out in Figure 2.6C, the diamond plants phyletic group. Such a group is called paraphyletic; a METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 15

(A) Pollen Inflorescence Figure 2.6 (A) The matrix from Figure 2.4C, but with charac- grooves Petals in heads ter states added for a conifer. (B) The unrooted network from Figure 2.4B, but with the conifer attached according to the char- Star plants < 3 free no acter states in Figure 2.6A. (C) The network of Figure 2.6B rooted with the conifer. Note that the evolutionary history is now the Circle plants 3 free no same as in Figure 2.5A.

Square plants 3 fused no

Diamond plants 3 fused yes

not not Conifer < 3 applicable applicable

(B) Inflorescence Pollen colpi Petals a head < 3 3 free fused no yes

(C)

yes

no Inflorescence fused a head free Petals Time 3

< 3 Pollen colpi

(A) Figure 2.7 Two different ways to draw the yes tree in Figure 2.6C. Note that the length does no not change, nor does the hypothesized order Inflorescence of events. a head fused Petals free Time 3

< 3 Pollen (B) colpi yes

Inflorescence a head no Time fused Petals

Pollen free 3 colpi

< 3 16 CHAPTER TWO paraphyletic group includes a common ancestor and network can be rooted to produce an evolutionary tree, some but not all of its descendants. cladogram, or phylogeny. Some taxonomic groups are cladistically unresolved; Two phenomena, however, make it much harder in they cannot be determined to be either positively para- practice to determine evolutionary history: parallelism phyletic or positively monophyletic, and are referred to and reversal, which sometimes are referred to together as metaphyletic. The way we have drawn the circle as homoplasy. Parallelism is the appearance of similar plants in Figure 2.5, for instance, indicates that we do not character states in unrelated organisms. (Many authors know whether they have a synapomorphy or not, and make a distinction between parallelism and conver- thus they form a metaphyletic group. gence, but for this discussion we will treat them as As mentioned earlier, a character state that is derived though they are the same.) A reversal occurs when a (synapomorphic) at one point in time will become ances- derived character state changes back to the ancestral tral later. In this example, tricolpate pollen is a shared state. To provide a clear example, divide the group that derived character of a large group of flowering plants. It we have called “star plants” into black star plants, gray is a synapomorphy and indicates monophyly of the star plants, and white star plants. Let us assume that the group sometimes called the eudicots. For the group with gray star and white star plants have only one cotyledon, fused petals, however, tricolpate pollen is an ancestral, whereas all the rest of the organisms have more than one or plesiomorphic character. It is something they all (including the conifer). Let us further assume that the inherited from their common ancestor and thus does not white star plants have fused petals. Add the character indicate relationship. Plesiomorphic similarities cannot cotyledon number to the matrix in Figure 2.6A to give show genealogical relationships in the group being stud- the matrix in Figure 2.8A, which gives the same informa- ied because they evolved earlier than any of the taxa tion as the network in Figure 2.8B. being compared, and merely have been retained in vari- Now we see that, according to this network, there ous lineages (taxa). have been two changes in petal fusion. Counting the It is sometimes possible to determine monophyly of a number of changes on this network (its length), we find group by observing that the characters do not occur in five: one each in pollen colpi, flowers in heads, and any other organism. For example, all members of the cotyledon number, and two in petal fusion. grass family (Poaceae) have an embryo that is unlike the A group based on fused petals would be considered embryo of any other flowering plant. We can thus polyphyletic. Polyphyletic groups have two or more hypothesize that the grass embryo is uniquely derived in ancestral sources in which the parallel similarities (is a synapomorphy for) the family and indicates that the evolved. (Although we distinguish here between para- family is monophyletic. This is the same as saying that phyletic and polyphyletic, many systematists have any reasonable rooting of the phylogenetic tree will lead observed that the difference is slight, and simply call any to the same conclusion. para- or polyphyletic group non-monophyletic.) Petal It is often possible to find evidence that a group is fusion in this case is nonhomologous since it fails the monophyletic even without a large computer-assisted ultimate test of homology— congruence with other char- phylogenetic analysis. Indeed, most cladistic analyses acters in a phylogenetic analysis. were done by hand until the mid-1980s. Characters are Why not draw the network in such a way that petal divided into character states, as with any cladistic analy- fusion arose only once? Such a network is shown in Fig- sis. The character state in the outgroup (or outgroups) is ure 2.8C. Now we have one change in petal fusion, but then assumed to be ancestral (Stevens 1980; Watrous and that requires two changes in cotyledon number, and also Wheeler 1981; Maddison et al. 1984). In other words, the two changes in number of pollen colpi, making the net- character is polarized, or given direction. The shared work six steps long. derived, or synapomorphic, state can then be used as Each of the networks (Figures 2.8B and C) can be con- evidence of monophyly, and cladograms can be con- verted to a phylogeny by rooting at the conifer, but they structed on the basis of shared derived character states. make different suggestions about how plants have This kind of thinking is often useful in providing a first evolved. In one case, cotyledon number and number of guess as to whether taxonomic groups might be mono- pollen colpi have been stable over evolutionary time, phyletic and thus named appropriately. whereas petal fusion has appeared twice, independently. In the other case, we postulate that cotyledon number CHOOSING TREES and number of pollen colpi have changed twice over As can be seen from the preceding sections, determining evolutionary time, while petal fusion has evolved only the evolutionary history of a group of organisms is con- once. By drawing either of these networks, we are mak- ceptually quite simple. First, characters are observed and ing a hypothesis about how evolution has happened— divided into character states. Second, using the character about which genetic changes have occurred, at what fre- states, a Venn diagram (Figure 2.4A), character × taxon quency, and in which order. matrix (Figure 2.4C), and branching network (Figure As you can see, the hypotheses differ. How do we 2.4B) can be constructed. Third, using an outgroup, the determine which is correct? There is no way to be certain. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 17

(A) Pollen Inflorescence Cotyledon Figure 2.8 (A) A plant by character matrix. (B) grooves Petals in heads number Unrooted network based on the matrix in Figure 2.8A. Note that petal fusion appears to change Black star plants < 3 free no 2 twice. Network length is 5. (C) Another possible unrooted network based on the matrix in Figure Gray star plants < 3 free no 1 2.8A. Unlike the network in Figure 2.8B, petal fusion changes only once, but cotyledon num- White star plants < 3 fused no 1 ber and pollen colpi change twice. Network length is 6. Circle plants 3 free no 2

Square plants 3 fused no 2

Diamond plants 3 fused yes 2

not not Conifer < 3 applicable applicable > 2

(B) Cotyledon Inflorescence Petals number Pollen colpi Petals a head fused free 1> 1< 3 3 free fused no yes

(C) Cotyledon Inflorescence number Pollen colpi Petals a head 1> 1 < 3 3 free fused no yes

> 1 Cotyledon 1 number 3 Pollen colpi < 3

No one was there to watch the evolution of these plants. Other methods use other optimality criteria. Instead of We can, however, make an educated guess, and some choosing the tree with the fewest evolutionary changes, guesses seem more likely than others to be correct. One one could convert the character matrix to a measure of way to proceed is to ask, “What is the simplest explana- similarity or dissimilarity among the plants, and then tion of the observations?” This rule, used throughout sci- build a network that minimizes the dissimilarity; this is ence, is known as Ockham’s razor: Do not generate a known as the minimum distance method. Alternatively, hypothesis any more complex than is demanded by the one could develop theories about the probability of data. Applying this principle of simplicity, or parsimony, change from one character state to another and then use leads us to prefer the shorter network. The fact that it is those probabilities to calculate the likelihood that a given shorter does not make it correct, but it is the simplest branching diagram would lead to the particular set of explanation of the data. data observed. The tree with the highest likelihood is pre- There are other ways to construct and to choose ferred—the maximum likelihood method (Felsenstein among evolutionary networks and trees. We have pre- 1981; Hillis et al. 1993; Huelsenbeck 1995; Swofford et al. sented a simple step-counting (parsimony) method here 1996). Maximum likelihood methods are particularly because it is the most widely used, the most easily suited to molecular data (see Chapter 5), for which it is applicable to morphological changes, and possibly also easier to model the probability of genetic changes (muta- the most intuitive method. Parsimony works well when tions). For a more comprehensive description of methods evolutionary rates are not so fast that chance similarities of phylogeny reconstruction, see Swofford et al. (1996). (due to the evolution of identical derived characters In the examples we have presented, in which there are independently in two or more lineages) overwhelm few characters and little homoplasy, it is easy to construct characters shared by the common ancestor. the shortest network to link the organisms. In most real 18 CHAPTER TWO cases, however, there are many possible networks, and it fore different histories for the same taxa. In addition, is not immediately obvious which one is the shortest. studies using different kinds of characters (e.g., gene Fortunately, computer algorithms have been devised that sequences, morphology) may find still other trees. compare trees and calculate their lengths. Some of the Rather than choosing among the trees in these cases, most widely used of these are PHYLIP (Felsenstein 1989), often systematists simply want to see what groups are hennig86 (Farris 1989), NONA (Goloboff 1993), and PAUP found in all the shortest trees, or by all methods of analy- 3.1.1/PAUP* 4.0 (Swofford 1993). These programs either sis, or among different kinds of character matrices. The evaluate data over all possible trees (an exhaustive information in common in these trees can be summa- search), or make reasonable guesses as to the topology of rized by the use of a consensus tree. the shortest trees (branch-and-bound searches or heuris- Strict consensus trees contain only those monophylet- tic searches). In analyses of numerous taxa, only heuristic ic groups that are common to all trees. For example, algorithms can be used. These algorithms may not suc- analyses of different sets of data have produced different ceed in finding the shortest tree or trees because of the results regarding the early evolution of the angiosperms. large number of possible dichotomous trees. For exam- A study of morphological characters and 18S rRNA ple, the possible interrelationships of three taxa can be sequences led to the evolutionary tree shown in Figure expressed by only three rooted trees, [A(B,C)], [B(A,C)], 2.9A (Doyle et al. 1994; we have omitted some taxa for the and [C(A,B)]. But the number of potential trees expands purposes of this example). A study of rbcL sequences led rapidly given larger numbers of taxa; for example, four to the tree in Figure 2.9B (Albert et al. 1994). For descrip- taxa yield 15 trees, five yield 105 trees, six yield 945 trees, tions of 18S rRNA and rbcL data, see Chapter 5. The trees and ten yield 34,459,425 trees! both show Gnetum as sister to Welwitschia, and those two as sister to Ephedra. (These three genera together make up SUMMARIZING EVOLUTIONARY TREES the gnetophytes; see Chapter 7.) Both trees also suggest Often parsimony analyses will find multiple trees, all that the Calycanthaceae and Laurales are closely related with the same length but with different linkages among (see Chapter 8). The strict consensus of the two clado- the taxa. Sometimes, too, different methods of analysis grams (Figure 2.9C) also shows the Gnetophytes and the will find trees showing different topologies and there- Calycanthaceae/Laurales clade.

(A) Morphology + 18S rRNA (B) rbcL

Gnetophytes Gnetum Welwitschia Ephedra Monocotyledons Piperales Eudicots Winteraceae Chloranthaceae Calycanthaceae Laurales Gnetum Welwitschia Ephedra Monocotyledons Piperales Eudicots Winteraceae Chloranthaceae Calycanthaceae Laurales

(C) Strict consensus (D) Semi-strict consensus Laurales Gnetum Welwitschia Ephedra Monocotyledons Piperales Eudicots Winteraceae Chloranthaceae Calycanthaceae Laurales Figure 2.9 (A) Phylogeny of angio- Gnetum Welwitschia Ephedra Monocotyledons Piperales Eudicots Winteraceae Chloranthaceae Calycanthaceae sperms based on data from morphol- ogy and 18S rRNA sequences (Based on Doyle et al. 1994). (B) Phylogeny of angiosperms based on data from rbcL sequences. (C) Strict consensus of trees in A and B. (D) Semi-strict con- sensus of trees in A and B. (Modified from Albert et al. 1994.) METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 19

There are differences between the two evolutionary DNA sequence, the starting point is usually a model in hypotheses, however. The rbcL tree suggests that the which mutation is assumed to be random, although this Piperales are sister to the monocots, but the morpholo- assumption is often modified to reflect hypothesized gy/rRNA tree tells us that the Piperales arose after the mechanisms of molecular evolution. The model is much monocot lineage diverged, such that the Piperales are more difficult for morphological characters, because we more closely related to the rest of the dicots. In Figure usually have no idea how many genes are involved, nor 2.9B, the eudicots, Winteraceae, and Chloranthaceae do we know what kinds of changes in the genes lead to appear as though they arose at the same time. This different character states. Nonetheless, certain assump- means that rbcL data cannot tell us whether they did tions must be made if one is to proceed at all. (And, we arise together, or one after the other, nor can we deter- note, there are no methods that are entirely free of mine the order. Having multiple lineages arising at the assumptions!) The major assumptions have to do with same apparent point in the diagram is usually an expres- the likelihood of particular changes of character states, sion of ambiguity. The difference in the position of the and the likelihood of reversals and parallelisms. Piperales combined with the ambiguity in the rbcL tree leads us to conclude that we really do not know which Ordering character states The characters in Figure early angiosperm lineages appeared first. This is reflect- 2.8A have only two states. Such two-state characters are ed in the strict consensus tree by drawing all those lin- interpreted as representing a single genetic switch— “on” eages as though they arose at the same time. producing one state (e.g., pollen is tricolpate), “off” When many trees are being compared, it is sometimes resulting in the other state (e.g., pollen is one-grooved, or interesting to know whether a clade appears in most of monosulcate). Over evolutionary time, of course, such the trees, even if it doesn’t occur in all of them. A majority- characteristics can continue to change. For example, tri- rule consensus tree can show all groups that appear in colpate pollen is modified in some Caryophyllales so that 50% or more of the trees. If a particular clade is present in it is spherical, with many pores evenly spaced around it the majority of the most-parsimonious trees, then this (looking rather like a golf ball); this pollen is pantoporate. clade will be represented on the majority-rule tree (along If we were to include the character “pollen grooves” in a with an indication as to the percentage of most-parsimo- matrix, it would now have three states—monosulcate, tri- nious trees showing that clade). The majority-rule con- colpate, and pantoporate. This is now a multistate char- sensus tree will be inconsistent with some of the original acter, in contrast to the two-state or binary characters trees, and thus provides only a partial summary of the discussed previously. Multistate characters create a phylogenetic analyses. dilemma: how many genetic switches are involved? A semi-strict, or combinable component, consensus It is possible that monosulcate pollen changed to tri- tree is often useful, particularly when comparing phylo- colpate, which then changed to pantoporate pollen, and genies with slightly different terminal taxa, or from dif- this actually matches what we think happened in evolu- ferent sources of characters. It is common, for example, tionary time (Figure 2.11A). (Recall that the outgroup to construct trees from two different sets of characters does not have tricolpate pollen.) This implies two genet- (e.g., a gene sequence and morphology) and to find that ic switches. It also implies that they must have occurred both sets of characters indicate monophyly of a particu- in order—pantoporate pollen could arise only after tri- lar group of species. Only one set of characters, however, colpate pollen. If we accept this series of events, the mul- may resolve relationships among the species. The semi- tistate character is considered to be ordered. If we decide strict consensus then indicates all relationships support- to allow for reversals of character states—that is, consid- ed by one tree or both trees and not contradicted by er the possibility that pantoporate pollen might switch either. For example, although the rbcL tree (Figure 2.9B) back to tricolpate and tricolpate to monosulcate pollen— does not give us information on the order in which eudi- the character is still ordered. It requires two evolutionary cots, Winteraceae, and Chloranthaceae originated, the (genetic) steps to go from monosulcate to pantoporate tree in Figure 2.9A does. The two trees are not really con- pollen, or two to go from pantoporate to monosulcate flicting; the morphology/rRNA tree just provides more pollen. A phylogenetic analysis in which all characters precise information. The semi-strict consensus thus fol- are treated as ordered is sometimes referred to in the lit- lows the morphology/rRNA arrangement of those three erature as Wagner parsimony. groups (Figure 2.9D). If we didn’t know anything about the plants involved, we might consider the possibility that monosulcate THE PROBABILITY OF EVOLUTIONARY CHANGE pollen might have changed to tricolpate pollen, and, in IN CHARACTERS an independent event, monosulcate pollen might have In trying to infer the evolutionary history of a group, we changed to pantoporate pollen (Figure 2.11B). This would depend on an implicit or explicit model of the evolution- suggest that there is a genetic switch from monosulcate to ary process. The more accurately the model reflects the tricolpate pollen and there is also a switch that allows underlying process, the more accurately we will be able change from monosulcate to pantoporate pollen, but a to estimate the evolutionary history. For nucleotides in a change from tricolpate to pantoporate pollen is impossi- 20 CHAPTER TWO

BOX 2A Long Branch Attraction

If there are great differences in the A B AB rates of character evolution between lineages such that some lineages are evolving very rapidly, and if the pat- Homoplasious characters tern of variation is sufficiently con- (parallelisms) strained (i.e., only a limited number of character states exist), then unusu- ally long branches tend to be con- C Homoplasious nected to each other whether or not character C they are actually closely related (Figure 2.10; Felsenstein 1978). This occurs because the numerous ran- dom changes, some of which occur in parallel in the two lineages, out- (A) True phylogeny: (B) Tree generated by parsimony number the information that shows 28 steps analysis: 26 steps common ancestry. The problem can- not be circumvented by acquiring Figure 2.10 Long branch attraction, a situation in which strongly unequal evolu- more characters; these would merely tionary rates cause parsimony to fail. (A) True phylogeny. Dotted lines show character add to the number of parallelisms states that have arisen in parallel in the lineages leading to A and B. (B) Phylogeny as linking the two rapidly evolving lin- reconstructed by parsimony.The number of parallelisms shared by A and B is greater eages. This situation, often called than the number of characters linking A and C, so A and B appear to be sister taxa, with long branch attraction or the parallelisms (in the true phylogeny) treated as shared derived characters of A and B. “Felsenstein zone,” can affect all methods of tree construction. With the correct model of evolution, how- determining the correct model may alleviated by including taxa that are ever, maximum likelihood methods be difficult). The situation basically is related to those terminating the long will not have this problem (although a sampling problem, and may be branches.

ble. The character in this case is still ordered, but in a dif- ferent way from Figure 2.11A. If reversals are possible, (A) then it requires two steps to get from tricolpate to panto- porate and two from pantoporate to tricolpate pollen. With morphological characters and character states, we are usually unsure of which switches are possible, so it is common to treat multistate characters as unordered (Fig- (B) ure 2.11C); this is sometimes called Fitch parsimony. In the case of an unordered character, we postulate only one switch between any two states. DNA sequence characters are multistate characters with four states (adenine, thymine, guanine, cytosine). To treat these as ordered would be nonsensical; adenine does not need to change to cytosine before changing to guanine. DNA characters are therefore always treated as unordered and fully reversible.

Reversals, parallelisms, and character weighting In the network in Figure 2.8B, we hypothesized that petal fusion arose twice, independently. To make the slight- (C) ly longer network in Figure 2.8C, we had to let cotyle-

Figure 2.11 Three alternative hypotheses about the evolution of pollen mor- phology. (A) Monosulcate changed to tricolpate, which then changed to panto- porate. As drawn, the character is ordered and irreversible. (B) Monosulcate changed to tricolpate and independently changed to pantoporate.The character is also ordered and irreversible. If the arrows were drawn as double headed, then the character would be interpreted as reversible. (C) Any pollen type can change to any other pollen type.The character is unordered and reversible. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 21 don number change from one to more than one and restriction site than to lose one (see Chapter 5). And com- back to one again—that is, to reverse. In comparing the plex characters (presumably controlled by many genes) trees in Figures 2.8B and 2.8C, therefore, we are com- may be weighted over simple characters (presumably paring the hypotheses that (1) mutations in the genes controlled by fewer genes), again because the latter are leading to petal fusion have happened more than once thought to be more labile over evolutionary time. versus (2) mutations in the genes controlling cotyle- The most common approach, used in most prelimi- don number have happened and then their effects nary analyses, is to consider all characters of equal have been reversed. In deciding that the network in weight. Although this sometimes is described as Figure 2.8B was shorter than the one in Figure 2.8C, we “unweighted,” in fact it assumes that all characters are counted all the steps equally, whether they were par- equally likely to change and weights them accordingly. allelisms, reversals, or unique origins. Underlying all discussion of weights is the assump- This may or may not be reasonable. Dollo’s law, for tion that characters of organisms evolve independently. example, suggests that for very complex characters, par- This assumption requires that change in one character allel origin is highly unlikely, whereas reversal may be does not increase the probability of change in another quite easy (Mayr and Ashlock 1991). The assumption is character. As with the previous assumption, this one that many genes must change to create a morphological may be violated frequently. For example, a change in structure, but only one of those genes needs to be modi- flower color may well lead to a shift in pollinators, fied to lose it. Dollo’s law can be built into the process of which would then increase the probability that corolla choosing a tree by making gains of structures count for shape would change. Violating this assumption obvious- more than losses; the process is then known as Dollo ly affects character weighting, in that the likelihood of parsimony. (To define the terms “gain” and “loss,” of change of two characters is not the same. course, requires a rooted tree; hence Dollo parsimony cannot be applied to an unrooted network.) DO WE BELIEVE THE EVOLUTIONARY TREE? Certain characters are sometimes weighted in cladistic An evolutionary tree is simply a model or hypothesis, a analyses. This reflects the assumption that certain charac- best guess about the history of a group of plants. It fol- ters should be harder to modify in evolutionary time lows that some guesses might be better, or at least more than others. One might hypothesize, for example, that convincing, than others. Much of the current literature leaf anatomy is less likely to change than leaf hairiness on phylogeny reconstruction involves ways of determin- (pubescence), and therefore a change in a leaf anatomical ing how much credence we should give to a particular character could be counted as equivalent to two changes evolutionary tree. Use of an optimality criterion is one in pubescence for the purposes of counting steps in the way to evaluate the evolutionary tree; of all possible tree. Such weighting decisions can easily become subjec- descriptions of history, we prefer the one that requires tive or arbitrary, and risk biasing the outcome of the the fewest steps, or gives the maximum likelihood, or study toward finding particular groupings. (For example, the minimum distance. It is usually possible to evaluate the investigator might theorize, “My favorite species trees more precisely, however. For the purposes of this group has interesting leaf anatomy; therefore I think that discussion, we will continue to focus on phylogenies leaf anatomy is phylogenetically important; therefore I generated according to maximum parsimony (i.e., the will give it extra weight in the phylogenetic analysis.” In fewest evolutionary steps). this case, it is no surprise when the favorite species group is shown to be monophyletic.) Measuring support for the whole tree: Assessing Because of the possibility of bias, systematists gener- homoplasy Parsimony analyses minimize the num- ally attempt to base weighting decisions on an objective ber of characters that change in parallel or reverse. If criterion. One approach is to do a preliminary phyloge- there are many such homoplasious characters, then it netic analysis with all characters assigned equal weights. is possible that the phylogenetic tree is an artifact of The results of this analysis will identify which characters the characters we have chosen, and a slight change in have the least homoplasy on the shortest tree(s); these characters will lead to a different tree. The simplest, characters with less homoplasy can then be given more and most common, measure of homoplasy is the con- weight in subsequent analyses, a process known as suc- sistency index (CI), which equals the minimum cessive weighting. amount of possible evolutionary change (the number Another approach is to base weights on knowledge of of genetic switches) divided by the actual tree length the underlying genetic basis of characters. For example, (the number of actual genetic changes on the tree). In transversions (purine → pyrimidine or pyrimidine → the network of Figure 2.4B, each of the three characters purine changes) are weighted over transitions (purine → represents a single genetic switch, and each one purine or pyrimidine → pyrimidine changes) because changes only once, so the consistency index is 3/3 = transitions are known to occur more frequently and be 1.0. In the network in Figure 2.8B, there are four bina- easier to reverse. Restriction site gains may be weighted ry (one-switch) characters, but one of them (petal over site losses because there are fewer ways to gain a fusion) changes twice on the tree, so that the consis- 22 CHAPTER TWO tency index is 4/5 = 0.80. Consistency indices may also instead of varying between 0 and 1, the CI in this case be calculated for individual characters and equal the varies between 0.5 and 1.0. The RI corrects for this nar- minimum number of possible changes (one, for a bina- rower range of the CI by comparing the actual number ry character) divided by the actual number of changes of changes in the character to the maximum possible on the tree. For example, the CI of petal fusion (Figure number of changes. 2.8B) is 1/2 = 0.50. For a given matrix (set of characters The RI is computed by calculating the maximum pos- and taxa), the shortest network or tree will also have sible tree length, which is the length that would occur if the highest consistency index. Lower consistency the derived character state originated independently in indices indicate many characters that contradict the every taxon in which it appears (i.e., if all taxa with the evolutionary tree. derived character state were unrelated). The minimum Comparing consistency indices across data sets is tree length and actual tree length are computed the same hazardous because the CI has some undesirable proper- way they are for calculating the CI. The RI then equals ties. For one thing, a character that changes once in only the maximum length minus the actual length, divided one taxon will have a consistency index of 1.0, but it in by the maximum length minus the minimum length, fact says nothing about relationships. Such a uniquely or (Max – L)/(Max – Min). In Figure 2.8B, then, the RI is derived character is sometimes called an autapomorphy. (9 – 5)/(9 – 4) = 4/5 = 0.80. For example, if one of the black star plants in Figure 2.8B had hairy leaves while all other plants studied had hairless ones, leaf hairi- 4 Cycad ness would not be of any help in indicating the relationship of the hairy-leaved plant. 2 Ginkgo The character is uninformative. Because the uninformative character changes only once, 1 Conifer however, it has a CI of 1.0. If we added many uninformative characters into the analysis, 3 Ephedra the overall CI would be inflated accordingly 10 and would give a misleading impression 3 Welwitschia that many characters supported the tree. 6 Uninformative characters, therefore, are 5 Gnetum often omitted before calculating the consis- 3 tency index. 5 Magnoliales The consistency index is also sensitive to the number of taxa in an analysis 6 5 Chloranthaceae (Sanderson and Donoghue 1989): analyses 5 with many taxa tend to have lower CIs 4 Laurales than analyses with fewer taxa. This occurs 4 2 with both molecular and morphological 3 Calycanthaceae data, and with analyses of species, genera, 12 or families. 3 4 Winteraceae Other measures are used to describe how characters vary over the tree. One of 3 Eudicots these, the retention index (RI), is designed 4 4 to circumvent another limitation of the CI 3 Trochodendrales (Forey et al. 1992; Wiley et al. 1991). The CI is designed to vary between near 0 (a char- 3 3 Aristolochiaceae acter that changes many times on the tree) and 1.0 (a character that changes only 0 Saururaceae once). But consider the plants in Figure 2.8 9 18 once more. For those described by the 3 Piperaceae matrix in Figure 2.8A, only two groups— the white star plants and the gray star 4 Nymphaeaceae plants—have a single cotyledon. If the sin- 4 gle-cotyledon plants form a clade, as in Fig- 2 Monocots ure 2.8B, then the CI for cotyledon number is 1.0. If they are unrelated, as in Figure Figure 2.12 Phylogeny of the angiosperms, based on data from Doyle et 2.8C, then the CI is 0.5 (1/2), which is the al. (1994). Numbers above branches indicate the number of characters chang- lowest possible value on the tree. Thus, ing along that branch. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 23

Measuring support for parts of trees With parsimo- angiosperms that share numerous characters that do ny methods, the shortest available tree is preferred over not change elsewhere on the cladogram are more one that is longer. It is possible, however, that some believable than a group that shares only a few highly parts of the tree are more reliable than others. This will homoplasious characters. occur if reversals and parallelisms (or simple misinter- Another way to assess how well the data support the pretation of characters) affect some groups of plants tree is to determine whether a group of interest occurs in more than others, or if there were very few evolutionary other trees that are almost equally short. Suppose, in changes in the history of a particular group. One simple other words, we ask if there are other ways to analyze way to evaluate this is to note the number of genetic the homoplasious characters that lead to trees that are changes that occur on the branch leading to a particular one, two, or three steps longer. group, along with the consistency indices of the charac- For example, in the tree shown in Figure 2.12, the ters. For example, one of the morphological trees pro- shortest trees indicate that the earliest diverging lineages duced by Doyle et al. (1994) found 18 changes on the in the angiosperms were the monocots and the water branch leading to the angiosperms (Figure 2.12), and of lilies (Nymphaeaceae; see Chapter 8). This implies that these 11 were in characters that had a CI of 1.0. In other the character of herbaceous stems is gained once and words, over half of the genetic changes that occurred then lost, whereas reducing the number of ovules per during the origin of the angiosperms produced novel carpel to one occurs only once, and oil cells are gained characteristics, found nowhere else. Groups like the once and lost once (Figure 2.13A). On the other hand,

(A) (B) Eudicots Eudicots Gnetum Gnetum Coniferales Welwitschia Ephedra Nymphaeales Monocots Piperaceae Saururaceae Aristolochiaceae Trochodendrales Winteraceae Magnoliales Calycanthaceae Chloranthaceae Laurales Coniferales Welwitschia Ephedra Magnoliales Laurales Calycanthaceae Chloranthaceae Winteraceae Trochodendrales Nymphaeales Monocots Piperaceae Saururaceae Aristolochiaceae Cycad Ginkgo Cycad Ginkgo

Ovules Lose 1 per carpel oil cells Lose oil cells Herbaceous Woody Ovules Gain oil cells 1 per carpel Herbaceous Gain oil cells

Shortest tree One step longer

(C) (D) Eudicots Eudicots Gnetum Gnetum Coniferales Welwitschia Ephedra Nymphaeales Monocots Piperaceae Saururaceae Aristolochiaceae Trochodendrales Winteraceae Magnoliales Calycanthaceae Chloranthaceae Laurales Coniferales Welwitschia Ephedra Magnoliales Laurales Calycanthaceae Chloranthaceae Winteraceae Trochodendrales Nymphaeales Monocots Piperaceae Saururaceae Aristolochiaceae Cycad Ginkgo Cycad Ginkgo

d = 1 d = 1 d = 1 d = 1 d = 1 d = 1 Shortest tree Strict consensus showing of shortest tree branches with and trees a decay value one step longer of one

Figure 2.13 (A) The same tree as in Figure 2.12, indicating patterns of change in pres- ence/absence of oil cells, ovule number per carpel, and plant habit. (B) An alternative tree, only one step longer than the tree in Figure 2.13A, showing patterns of change in the same characters. Note that herbaceousness now is hypothesized to have evolved only once, but loss of oil cells and reduction of ovule number occur twice. (C) Strict consensus of the shortest trees and trees one step longer (Figures 2.13A and B). (D) The same tree as Figures 2.12 and 2.13A, showing branches with a decay value of one. (Data from Doyle et al. 1994.) 24 CHAPTER TWO trees one step longer, in which the earliest angiosperm character (inflorescence a head) was missed by the ran- lineages led to the magnolias, suggest that herbaceous dom selection process. Multiple such randomized matri- stems evolved once, but reduction in ovule number ces are constructed, and the most-parsimonious tree(s) occurred twice, and there were three changes in oil cells found for each new matrix. This leads to a set of at least (gained once and lost twice or vice versa) (Figure 2.13B). 100 trees, which can be summarized by a consensus tree Thus, by looking at trees one step longer, some charac- (see pages 18–19). In the bootstrap consensus tree, a clade ters are hypothesized to be less homoplasious, but some with a bootstrap value of, say, 95% was present in 95% of to be more so. If we now take the strict consensus of all the cladograms generated in the bootstrap analyses. the trees, including the shortest ones and those one step An example of a cladistic analysis giving both boot- longer, all the early angiosperm lineages are drawn as strap and decay values (along with branch lengths) is though they radiate from a single point, indicating represented in Figure 2.15. We see that bootstrap and uncertainty about the order in which they evolved (Fig- decay values are high for the genus Lyonia, indicating ure 2.13C). In other words, many of the branches that are that the data support monophyly of the genus, whereas evident in the shortest trees do not appear in trees one the linkage of Agarista and Pieris is supported by only step longer. Thus all those branches are not drawn in the 51% of the bootstrap trees, and in trees only one step strict consensus; they “collapse.” This can be indicated longer the two genera are not sisters, indicated by the by placing a one next to each of the collapsing branches notation d (decay) = 1. of the shortest tree (Figure 2.13D). The number is the Another excellent way to gain confidence in the decay index, sometimes called Bremer support, which groupings present in a tree is to compare phylogenies represents how many extra steps are required to find that have been based on different sets of characters. For trees that do not contain a particular group. It provides a example, phylogenies based on morphology, chloroplast relative measure of how much the homoplasy in the DNA nucleotide sequences (cpDNA), and nuclear DNA data affects support for a particular group. nucleotide sequences could be (and often are) compared. The decay index is not statistical, which, depending If these phylogenies show similar groups, then we can on one’s point of view, is either a virtue or a drawback. be more confident that they reflect the true order of Because history, and therefore the phylogeny, hap- events. For example, the monophyly of such families as pened only once and cannot be repeated, it is impossi- the Poaceae, Onagraceae, Ericaceae, Asteraceae, and ble to replicate the evolutionary experiment. It is cer- Orchidaceae has been supported by phylogenetic analy- tainly possible, however, to test whether character data sis of many kinds of data, including information from are different from random, although there are many morphology, chloroplast gene sequences, and nuclear possible ways to randomize systematic data. Many gene sequences. tests have been devised that use some sort of random- Comparing trees is often particularly intriguing when ization technique. Probably the most widely used is the data come from different genes; a more extensive dis- bootstrap analysis. cussion of this is in Chapter 5. It is also common to com- Bootstrap analysis randomizes characters with bine morphological and DNA characters in a single phy- respect to taxa. As an example, begin with the matrix in logenetic analysis, which often leads to more strongly Figure 2.8A and randomize the columns while leaving supported phylogenies than either sort of data can pro- the rows in place. Choose a column at random to become duce alone. the first column of the new matrix. Then choose another column from the original matrix to become the second column, and so on until a new matrix is created with the same number of columns as the original. Because one Pollen Pollen Cotyledon returns to the original matrix each time to choose a new colpi Petals colpi number column, some characters may be represented several times in each new matrix, while others are omitted. This Black star plants < 3 free < 3 2 is usually described as random sampling with replace- Gray star plants < 3 free < 3 1 ment. Thus, Figure 2.14 shows the matrix in Figure 2.8A sampled with replacement; note that the first character White star plants < 3 fused < 3 1 (pollen colpi) has been selected twice, whereas the third Circle plants 3 free 3 2

Square plants 3 fused 3 2 Figure 2.14 The matrix from Figure 2.8A sampled with replacement, as it would be for the first step of a bootstrap Diamond plants 3 fused 3 2 analysis. Note that in the sampling process, the character not “pollen colpi” has been sampled twice, whereas the character Conifer < 3 < 3 > 2 “inflorescence a head” has been omitted. applicable METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 25

17 Figure 2.15 The single most parsimonious 4(90) Leucothoe racemosa tree found in branch-and-bound analysis of 15 the Lyonia group (taxa in boldface type) using 22(100) Gaultheria eriophylla matK data. Branch lengths appear above lines; d > 6 (100) 20 12 Satyria warszewiczii bootstrap values are in parentheses; decay 8 index (d) is below lines. Length = 425, consis- d > 6 Vaccinium macrocarpon tency index 0.60. (From Kron and Judd 1997.) 9 7(97) Pieris formosa 10 5(94) d = 6 Pieris phillyreifolia Pieris d = 4 6 12(98) 6(88) Pieris floribunda d = 4 d > 6 16 2(51) Pieris nana d = 1 14 3 Agarista populifolia 4(67) Agarista d = 1 12 d = 1 Agarista salicifolia 21 (73) Craibiodendron yunnanense 8 17 d = 2 9(97) Lyonia ovalifolia 4 7(55) d = 5 Lyonia ligustrina d = 2 Lyonia 18(100) Lyonia lucida d > 6 Lyonia ferruginea 19 35(100) Sphenotoma dracophylloides 22 d > 6 Sprengelia incarnata 25 Harrimanella hypnoides

Describing Evolution: many other uses. If a phylogeny is to be used to describe Mapping Characters on Trees history, however, it requires careful attention to the char- acters and character states used in the description. In Once created, phylogenies may be used as the basis of what follows we will focus on morphological characters, classification. This is one major goal of systematics and is but many of the points apply to any sort of characters. described in detail in the next section. Phylogenies can Consider a group of plants for which the tree is also be used to describe the evolutionary process and to known; a good example is the Ericaceae, for which much develop hypotheses about adaptation, morphological information is available (Figure 2.16). Assume for the and physiological change, or biogeography, among purposes of this discussion that this tree is an accurate

BOX 2B Phylogenetic Analysis Assumes That Evolution Can Be Diagrammed as a Branching Tree

Phylogenetic studies assume that hybrids, which have reticulating relationships among nonhybrid taxa. after two lineages diverge from evolutionary histories. However, recent studies by McDade each other, they never exchange Interspecific hybridization is (1990, 1992, 1997) indicate that genetic information again. This known to be common in plants, and hybrids are unlikely to create prob- assumption may in fact be violated the proper treatment of hybrids in lems in phylogenetic analysis unless frequently. If hybridization is com- cladistic analyses has been much dis- they are between distantly related mon, a plant may share the derived cussed (Bremer and Wanntorp 1979; parental species. When hybrids are characters of two unrelated parent Bremer 1983; Wagner 1980, 1983; recognized and their ancestry deter- plants, and the history will look Funk 1985; Kellogg 1989; Kellogg et mined (see Chapter 6), they can be more like a piece of macramé than al. 1996). Most systematists have manually inserted into the clado- like a tree. Phylogenetic analysis suggested that hybrids be identified gram, which then indicates not only will always produce a treelike dia- and removed from analyses because cladogenetic events (brought about gram, whether appropriate or not. their inclusion could lead to through speciation) but also reticu- Phylogenetic methods presuppose increased homoplasy, an increased lating histories (developed through divergent evolution and cannot number of most-parsimonious trees, interspecific hybridization). give the correct phylogeny for and a distortion of the patterns of 26 CHAPTER TWO reflection of history, and that each of the terminal genera really is monophyletic, which has been demonstrated by studying multiple species of each. Then consider a study 1 2 3 that is concerned with the gain or loss of fused petals, Oxycoccum

which are intimately connected with the evolution of sect. pollination systems. This is the kind of study that sys- Vaccinium tematists frequently engage in, because the details of Harrimanella Empetrum Lyonia Rhododendron Rhododendron Epacris Rhododendron Cassiope Enkianthus Ledum character evolution lead to hypotheses about how natur- Vaccinium al selection has worked. Also, when constructing classifi- Fused: other Fused: Free: Fused: Free: Fused: Fused: Fused: Fused: Fused: Fused: cations, one frequently wants to know what morpholog- Fused: ical characters can be attributed to and distinguish a particular monophyletic group. In Figure 2.16, we show the observed character states for the genera. It seems trivially obvious from looking at the distribution of characters and character states that free petals must have evolved in the lineage leading to Ledum (Labrador tea) and again in the lineage leading to Vaccinum sect. Oxycoccum (cranberries). Phrasing this another way, the ancestor of Vaccinium sect. Oxycoccum and all other vacciniums (blueberries) had fused petals, ▲ as did the ancestor of Ledum plus Rhododendron sect. 3. Figure 2.16 Phylogeny of a portion of the Ericaceae, based Examine this “obvious” conclusion more closely. If we on data summarized in Stevens (1998).The genus were studying only species of Vaccinium, we would have Rhododendron is paraphyletic and is represented by three sep- no way of knowing whether fused petals were ancestral or arate lineages, numbered one to three.Two changes to free derived (Figure 2.17A). There must have been one genetic petals are hypothesized. change, but it could as easily have happened in the lineage leading to the cranberries (sect. Oxycoccum) as in the lin- eage leading to the blueberries. It is only by reference to the outgroup Epacris that we can determine when petal fusion was lost. Because Epacris has fused petals, free petals must have originated within Vaccinium; it is simplest (most par- (A) simonious) to assume just one genetic change, from fused to free (Figure 2.17B). This is the same as saying that the Oxycoccum Oxycoccum ancestor of blueberries plus cranberries had fused petals. If sect. we were to postulate that the ancestor had free petals, then sect. Vaccinium we would need two changes to fused petals—one in Vaccinium Epacris and one in the blueberries. The same argument or Vaccinium applies in the case of Rhododendron and Ledum. Vaccinium Now suppose that we were studying only species of Free: Fused: other Fused: other Vaccinium, but this time, instead of using Epacris or other Free: Ericaceae as outgroups, we used only Ledum. This could easily happen if material of the other genera were hard Ancestor fused Ancestor free to obtain, or if they were extinct and we didn’t even know they had existed. Now we would conclude that the ancestor of all vacciniums had free petals, and that in (B) response to some unknown selective pressure there was a change to fused petals (Figure 2.18A). This is exactly the Oxycoccum opposite conclusion from the one reached above, and the only Oxycoccum sect. difference is the genera included in the analysis. sect. Vaccinium Vaccinium ▲ Epacris or Epacris Vaccinium Figure 2.17 (A) Two taxa differ in character states. It is Vaccinium impossible to determine from this information alone what the Free: Free: Fused: other Fused: other Fused: character state of the ancestor was because either assumption Fused: will involve one change in one descendant lineage. (B) The addition of an outgroup determines the character state of the ancestor. In this case, it is simpler (requires fewer steps) to assume that the ancestor had fused petals. Ancestor fused Ancestor free METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 27

(A) Figure 2.18 (A) Analysis of character state change in Vaccinium using a different outgroup. Note that the inference of the ancestral state is exactly the opposite of that reached using Epacris. (B) Analysis of char- acter state change in Vaccinium using two outgroups that differ in state. Oxycoccum Oxycoccum It is now impossible to determine the character state of the ancestor. sect. sect. Vaccinium Vaccinium

or Ledum Vaccinium Vaccinium Ledum to postulate that the ancestor had free petals and there Free: Free: Fused: other Fused: other Free: Fused: were two changes to fused. These two choices are known as equally parsimonious reconstructions. It is safe to say that for many characters on many trees, there Ancestor fused Ancestor free are multiple equally parsimonious reconstructions. In other words, there are multiple equally good hypothe- ses about the direction and timing of character state (B) change. If you return to the example in Figure 2.13, you should be able to find equally parsimonious reconstruc- tions that differ from the ones shown. Oxycoccum Oxycoccum Ambiguity can also come from including taxa for sect. sect. which the character state is not known. Suppose, for Vaccinium Vaccinium example, two new taxa are discovered such that, on the basis of other characters, one is clearly sister to Vaccinium or Vaccinium Vaccinium Ledum Ledum sect. Oxycoccum, and the other sister to the rest of Vaccini- um (Figure 2.19). In addition, suppose that it is unclear Free: Free: Free: Free: Fused: other Fused: other Fused: Rhododendron 3 Fused: Rhododendron 3 whether the petals are fused or free. (This is more com- mon that you might think; it can occur when the original description is vague and/or illustrations are unclear, or Ancestor Ancestor when the original plant is known only from fruiting fused free material.) This now means that we do not know what the ancestral state was for Vaccinium, so that we cannot make any hypothesis about direction of evolutionary One might try to improve the situation by using change. It also means that we cannot be sure that fused additional outgroups. For example, consider the same petals is a synapomorphy for the genus. study of Vaccinium, but now use both Ledum and Rhodo- Various algorithms have been developed to assign dendron as outgroups. In this case, the direction of character state changes to particular portions of trees (see change is completely ambiguous (Figure 2.18B). It is as Chapter 5 of Maddison and Maddison 1992 for a lucid simple to postulate that the ancestor of the group had and comprehensive discussion of these). Depending on fused petals and there were two changes to free as it is the algorithm used, the character changes can be biased in favor of parallelisms (the so-called “delayed transfor- mation,” or DELTRAN algorithm) or in favor of revers- als (“accelerated transformation,” or ACCTRAN). The results can have implications, sometimes major, for hypotheses about the evolutionary process, and may also Oxycoccum affect how organisms are described in a classification. sect. Vaccinium Constructing a Classification

Vaccinium The theory of classification is a topic with which system- atists have been wrestling for centuries, leading to a Fused: other New species 1 New species 2 Free: ?? broad and frequently contentious literature (see Chapter 3). The principles of phylogenetic classification outlined Petals fused Petals fused or free or free here are commonly but not universally held. In general, however, there are several goals of classification. A clas- Petals fused sification is a common vocabulary designed to aid com- or free munication. A classification should be stable; names that Figure 2.19 Addition of species for which the character state are frequently changed become useless for communica- is unknown can prevent any inference about the ancestral state. tion. A classification should be predictive; if you know 28 CHAPTER TWO the name of a plant, it should help you to learn more entire group of plants with tricolpate pollen (circle plants about it, and guide you to its literature. plus Asteridae) is monophyletic and is known as the Systematists generally agree about the goals of classi- eudicots (or the tricolpate clade). This group could be fication, but may disagree profoundly on how to reach given a formal Latin name, but it does not have one at those goals. In this text, we take a particular point of the moment. view, using phylogenetic classifications throughout. In cladistic classification, paraphyletic groups are not Thus, to the greatest extent possible, we have employed named. In Figure 2.6C, a group made up of square monophyletic and avoided paraphyletic or polyphyletic plants plus circle plants would be paraphyletic. The groups. In the few cases where a non-monophyletic fam- most recent common ancestor shared by any square ily or order has not yet been divided into monophyletic plant and a circle plant (dots on the diagram) is also the units, we have placed the taxon name in quotation most recent common ancestor of any square plant and a marks. The monophyly of many genera of angiosperms diamond plant. In other words, the square plants are as is questionable, but so few phylogenetic analyses are distantly related to circle plants as any one of them is to available at this level that possible or probable paraphy- diamond plants. If we name a group that included the ly or polyphyly of genera is not indicated. square plus the circle plants, it would imply that the two The biological diversity on Earth is the result of plants are closely related, whereas they are not. genealogical descent with modification, and mono- There are many examples in this book of named phyletic groups owe their existence to this process. It is groups of plants that we now believe to be paraphyletic. appropriate, therefore, to use monophyletic groups in One well-known example is “bryophytes,” often used to biological classifications, so that we may most accurately refer to the the non-vascular land plants (liverworts, reflect this genealogical history. Classifications based on hornworts, and mosses; see Figure 1.1). But the liver- monophyletic groups will be more predictive and of worts, hornworts, and mosses are more distantly related greater heuristic value then those based on overall simi- to each other than the mosses are to the vascular plants larity or idiosyncratic weighting of particular characters (tracheophytes). If we refer to bryophytes (without quo- (Donoghue and Cantino 1988; Farris 1979). Phylogenetic tation marks), the name implies a closer relationship classifications, because they reflect genealogy, will be the than actually exists. most useful in biological fields, such as the study of Several traditionally recognized plant families are plant distributions (phytogeography), host/parasite or paraphyletic; for example, Apocynaceae and Cappa- plant/herbivore interactions, pollination biology, and raceae. In this text, these have been recircumscribed so as fruit dispersal, or in answering questions related to the to recognize monophyletic groups: Apocynaceae have origin of adaptive characters (Brooks and McLennan been combined with Asclepiadaceae, and Capparaceae 1991; Forey et al. 1992; Humphries and Parenti 1986; with Brassicaceae. Nelson and Platnick 1981). Because of its predictive framework, a phylogenetic classification can direct the NAMING: NOT ALL GROUPS ARE NAMED search for genes, biological products, biocontrol agents, A phylogenetic classification attempts to name only and potential crop species. Phylogenetic information is monophyletic groups, but the fact that a group is mono- also useful in conservation issues. Finally, phylogenetic phyletic does not mean it needs to have a name. The rea- classifications provide a framework for biological sons for this are practical. We could put every pair of knowledge and the basis for comparative studies linking species into its own genus, every pair of genera into its all fields of biology (Funk and Brooks 1990). own family, every pair of families into its own superfam- Constructing a classification involves two steps, the ily, etc. Such a classification would be cumbersome; it first being the delimitation and naming of groups. In a also would not be stable, because our view of sister phylogenetic classification this is uncontroversial: species would change each time a new species is named groups must be monophyletic. The second step described, and our view of the entire classification involves ranking the groups and placing them in a hier- would have to shift accordingly. In practice, there are archy. This remains problematical. many monophyletic groups that are not named. For example, the genus Liquidambar (sweet gum) is mono- GROUPING: NAMED GROUPS ARE MONOPHYLETIC phyletic and contains four species. Although the rela- A phylogenetic classification reflects evolutionary histo- tionships among the four species are quite clear, the ry and attempts to give names only to groups that are pairs of species are not named, and few systematists monophyletic—that is, an ancestor and all its descen- would consider doing so. In another example, over half dants. In the example in Figure 2.6C, we infer that the of the genera of the grass family fall into a single large Asteraceae (diamond plants) are monophyletic because clade which contains four traditionally recognized sub- they have flowers in heads. The square plants plus families. Although agrostologists refer to this clade as Asteraceae are also monophyletic because they share the the PACC clade (an acronym for Panicoideae-Arundin- derived character state of fused petals; this group has a oideae-Centothecoideae-Chloridoideae), it has no formal name, the Asteridae (or the asterid clade). Similarly, the Latin name. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 29

How do systematists decide which monophyletic A fourth criterion is nomenclatural stability. A classifi- groups to name? There is no codified set of rules, but cation is ultimately a vocabulary, a means of communi- several criteria have been suggested by various authors, cation. It cannot function this way if the meanings of the and some criteria are in common use despite not being names continually change. Thus given a set of well-sup- fully articulated. A major criterion—perhaps the major ported, diagnosable, monophyletic groups, ones that criterion—is the strength of the evidence supporting a have been named in the past can—and we would argue group. Ideally, only clades linked by many shared should—continue to be named. This is yet another argu- derived characters should be formally recognized and ment against formally naming the PACC clade of the named in classifications. This makes sense if a classifica- grasses, in that it would entail an unnecessary set of tion is to function as a common vocabulary. Names are changes affecting long-standing taxonomic usage. It is most useful if they can be defined, and the more precise also an argument against dividing Liquidambar into two the definition the better. In other words, if a clade is to be genera, even though both would be monophyletic and named, it should have some set of characters by which it well-supported; both size of group and nomenclatural can be distinguished from other clades, or diagnosed. stability argue against the division (Backlund and Bre- This also relates to nomenclatural stability. If the mean- mer 1998; Stevens 1998). ing of a name shifts every time a new phylogeny is pro- duced or a new character is examined, then the name RANKING: RANKS ARE ARBITRARY becomes effectively meaningless. Having decided which monophyletic groups to name, A second criterion is the presence of an obvious mor- there is still the question of exactly how to name them. phological character. Although systematists are not like- The groups could, for example, be numbered, and a cen- ly to agree on the importance of this criterion, it is an tral index could list what is encompassed by the num- important extension of the idea of a well-supported bered group. This is similar to the system used by the tele- group, and is also relevant to the use of classifications by phone company to organize telephones. The difficulty, of non-systematists for identification purposes. If, for course, is that without a telephone book (a central index) example, the only way a field biologist can identify an and/or an excellent memory the system is inaccessible. organism is by knowing whether it has an alanine or a Biological classification attempts to provide a working serine at position 281 in its ribulose 1,5 bisphosphate car- vocabulary that conveys phylogenetic information, yet boxylase/oxygenase molecule, she may not find the clas- can be learned by biologists who are not themselves pri- sification of much help in making predictions about the marily systematists. Because a phylogeny is similar in organism. If, on the other hand, she knows that the structure to a hierarchy, in which small groups are includ- organism is a grass with a particular spikelet structure, ed in larger groups, which themselves are included in still then she can easily and reliably infer many aspects of its larger groups, it makes sense to reflect it as a hierarchy. biology. (Lack of an obvious morphological synapomor- Botanical classification uses a system developed in the phy is one of several reasons that the PACC clade of the eighteenth century, in which taxa are assigned particular grasses is not given a name.) The characters used for ranks, such as kingdom, phylum, class, order, family, classification do not have to be those used for identifica- genus, and species (i.e., Linnaean ranks; see Chapter 3 tion, but many systematists prefer to name clades that and Appendix 1). A classification of named monophylet- are easily recognized morphologically. ic groups should be logically consistent with the phyloge- Another criterion is size of the group. Human memo- netic relationships hypothesized for the organisms being ry is easily able to keep track of small numbers of items classified (as expressed in the sequence of branching (in the range of 3–7; Stevens, 1998), but to organize and points in the cladogram). That is, the categorical ranks of remember larger number of items requires additional a Linnaean classification can be used to express sister- mnemonic devices. (As an example of this, consider how group relationships. It is important to realize that, many 9-digit zip codes you can remember compared to although monophyletic taxa are considered to represent the 5-digit variety, or to 7-digit telephone numbers.) real groups that exist in nature as a result of the historical Dividing a large group into smaller groups is a way to process of evolution, the categorical ranks themselves are organize one’s thinking about large numbers of taxa. In only mental constructs. They have only relative (not the words of Davis and Heywood (1963), “We must be absolute) meaning (Stevens 1998). In other words, the able to place taxa in higher taxa so that we can find them familial level is less inclusive than the ordinal level and again.” The genus Liquidambar could be redefined to more inclusive than the generic level, but there are no cri- include only Liquidambar styraciflua and L. orientalis, and teria available to tell one that a particular taxon, such as a new genus could be described to include L. acalycina the angiosperms, should be recognized at the level of and L. formosana. There seems little reason to do this, phylum, class, or order. however, because four species is not a difficult number In Figure 2.20, a cladogram of imaginary taxa A–E is to keep track of. That said, there seems little reason to first converted into a hierarchical classification using Lin- divide a large group if well-supported clades cannot be naean ranks. Note that subgenus DE is nested within identified within it. genus CDE, which is, in turn, nested within family 30 CHAPTER TWO

ABCDEFigure 2.20 Alternative classifications based on the phylogeny of a hypothetical group of taxa ABCDE. One classification uses only three ranks (family, genus, species) plus a sequencing convention, whereas the other uses four ranks (family, genus, subgenus, and species).

es of stability however, it makes sense to leave the two genera in Poaceae, where they have been given a subfa- milial name, the Anomochlooideae. For more discussion of the problems encountered in Categorical rank Sequencing convention using the Linnaean system in phylogenetic classification, Family ABCDE Family ABCDE students should consult de Queiroz and Gauthier (1990, Genus AB Genus AB Species A Species A 1992); Forey et al. (1992); Wiley (1981); Wiley et al. (1991); Species B Species B and Hibbett and Donoghue (1998). Some systematists Genus CDE Genus C have proposed abandoning the Linnaean system alto- Subgenus C Species C Species C Genus DE gether and replacing it with a “phylogenetic taxonomy” Subgenus DE Species D in which monophyletic groups would be given unranked Species D Species E names, defined in terms of common ancestry, and diag- Species E nosed by reference to synapomorphies (de Queiroz and Gauthier 1990, 1992). Full exploration of this possibility is beyond the scope of this text. ABCDE. (But we could have treated clade ABCDE as an order, with clade CDE as a family and clade DE as a COMPARING PHYLOGENETIC CLASSIFICATIONS genus.) These procedures often lead to difficulties because WITH THOSE DERIVED USING OTHER TAXONOMIC in order to fully express the sister group relationships (in METHODS the cladogram), one needs more ranks than are available Not all taxonomists use phylogenetic methods, although (in the taxonomic hierarchy). Although additional ranks this is the majority approach. Some systematists have can be created by use of the prefixes super- and sub-, these held the view that, although evolution has occurred, par- may still be insufficient. Therefore, modifications to the allelism and reversal are so common that the details of method of classification outlined above have been pro- evolutionary history can never be deciphered. This point posed (Wiley 1979, 1981), such as the sequencing conven- of view led to a school of systematics known as phenet- tion, which states that taxa forming an asymmetrical part ics. Pheneticists argued that, since evolutionary history of a cladogram may be placed at the same rank and could never be unequivocally detected, organisms might arranged in their order of branching. The sequence of best be classified according to overall similarity. Thus, names in the classification denotes the sequence of similar organisms were placed together in a group, branching in the cladogram. Note that this is the same as while very different organisms were placed in different saying that not all monophyletic groups are given names. groups (Sneath and Sokal 1973). Even though ranking is arbitrary, the criteria One serious difficulty with the phenetic point of view described above for deciding which groups to name can was that many systematists produced treelike diagrams also be applied to deciding at what level to rank a group that grouped organisms by overall similarity, but these dia- (see Stevens 1998 for full discussion). Nomenclatural sta- grams were then interpreted as though they reflected evo- bility again becomes important. Often one of the mono- lutionary history. Sometimes this led to results similar to phyletic groups that could be given the name of family already has a commonly used family name, so it makes sense to continue to use the name family for these taxa. For example, it has recently been shown that the earliest diverging lineage in the Poaceae includes only two All other Poaceae extant genera, Anomochloa and Streptochaeta, so that the Anomochloa Streptochaeta phylogeny looks like that in Figure 2.21. One could, in principle, create a new family for Anomochloa and Strep- tochaeta; after all, it would be monophyletic and would leave the Poaceae as also monophyletic. For the purpos-

Figure 2.21 Phylogeny of Poaceae, showing the position of the genera Anomochloa and Streptochaeta. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 31

(A) Map Figure 2.22 Two graphical means of expressing phenetic relationships. (A) Maplike diagram. (B) Phenogram. Species I 7

3 Species III the taxa that were most similar, with the similarity rela- tionships illustrated on either a maplike or treelike dia- 6 gram (a phenogram; Figure 2.22). Phenograms were con- Species II structed using clustering algorithms, while maplike Numbers = Character differences diagrams resulted from ordination studies employing multivariate statistical procedures (see Abbot et al. 1985). Phenetic methods were used to produce classifica- (B) Phenogram tions, many of which are useful for identification and I II III information retrieval. These classifications were not 0 designed to retrieve evolutionary history, however, and are thus not appropriate for asking evolutionary ques- 1 tions. Phenetic systems do not distinguish between 1.5 synapomorphy and convergent or parallel evolution. 2 Evolutionary taxonomy differed from phylogenetic taxonomy in its approach to classification. The morpho- logical similarity of a group was of utmost importance, 3 and monophyly and paraphyly (in the strict cladistic Number of senses of those words) were secondary. Thus a group character 4 differences could be recognized on the basis of some combination of derived and ancestral, unique and shared characters 5 (Figure 2.23). Importance was given to the recognition of “gaps” in the pattern of variation among phylogenetical- ly adjacent groups (Simpson 1961; Ashlock 1979; Cron- 6 6.5 quist 1987; Mayr and Ashlock 1991). Characters consid- ered to be evolutionarily (or ecologically) significant 7

8 DBA1 E 10 C 3 11 those produced by a phylogenetic analy- sis, but sometimes it led to the production of “groups” made up of organisms that Autapomorphies, shared only the fact that they were differ- used to separate Monophyletic A from CB ent from everything else, including each group, other. Such groups have since proven to recognized Paraphyletic group, be paraphyletic or polyphyletic. because recognized because separated phenetically uniform The development of phenetic meth- from CB ods was an important prelude to the and A by acceptance and use of phylogenetic “gap” in pattern of approaches. A taxonomist constructing a variation phenetic classification would first care- fully observe as many characters as pos- sible. These characters were divided into states, or the quantitative value of the character merely Classification would be recorded (for example, a series of measure- Family ABCDE Figure 2.23 Phylogeny and ments of leaf length, with the mean recorded for each Genus DE a resulting non-phylogenetic taxon). This information was arranged in a taxon by Species D classification produced accord- character matrix similar to that in Figure 2.8A. This Species E ing to the evolutionary school. Genus CB matrix was converted to a similarity matrix (or taxon × The classification includes a Species C mix of monophyletic and taxon matrix) using any of several mathematical mea- Species B paraphyletic groups, separated sures of similarity (or dissimilarity; see Sneath and Sokal Genus A from each other by morpho- Species A 1973; Abbot et al. 1985). The systematist then grouped logical“gaps.” 32 CHAPTER TWO were stressed, and the expertise, authority, and intuition might define art and science), in part because so many of individual systematists were considered to be signifi- aspects of the discipline seemed to have no objective cant. Finally, although evolutionary classifications usual- basis. One fortunate result of phylogenetic systematics ly referred to evolution, and the groups recognized in is that at least one major aspect—the delimitation of such classifications were often called monophyletic, the groups—has become formalized such that there is gen- taxa were expected to be morphologically homoge- eral agreement on how it should be done. Whereas neous, and to be separated from each other by discrete phenetic and evolutionary classifications were ambigu- gaps (Ashlock 1979; Mayr and Ashlock 1991; Stevens ous about grouping criteria, phylogenetic classifica- 1986; Stuessy 1983, 1990). tions are precise. A named group can be taken as mono- It has been said that systematics is as much an art as phyletic, including all descendants of a single common a science (although this begs the question of how one ancestor.

Literature Cited and Suggested Readings

Abbot, L. A., F. A. Bisby and D. A. Rogers. de Queiroz, K. and J. Gauthier. 1992. Phylo- Gift, N. and P. F. Stevens. 1997. Vagaries in the 1985. Taxonomic analysis in biology. genetic taxonomy. Annu. Rev. Ecol. Syst. 23: delimitation of character states in quantita- Columbia University Press, New York. [An 449–480. tive variation: An experimental study. Syst. introduction to various phenetic methods.] *Donoghue, M. J. and P. D. Cantino. 1988. Biol. 46: 112–125. Albert, V. A., A. Backlund, K. Bremer, M. W. Paraphyly, ancestors, and the goals of tax- *Givnish, T. J. and K. J. Sytsma. 1997. Chase, J. R. Manhart, B. D. Mishler and K. onomy: A botanical defense of cladism. Homoplasy in molecular vs. morphologi- C. Nixon. 1994. Functional constraints and Bot. Rev. 54: 107–128. cal data: The likelihood of correct phyloge- rbcL evidence for land plant phylogeny. Doyle, J. A., M. J. Donoghue and E. A. netic inference. In Molecular evolution and Ann. Missouri Bot. Gard. 81: 534–567. Zimmer. 1994. Integration of morphologi- adaptive radiation, T. J. Givnish and K. J. Ashlock, P. D. 1979. An evolutionary systema- cal and ribosomal RNA data on the origin Sytsma (eds.), 55–101. Cambridge Univer- tist’s view of classification. Syst. Zool. 28: of angiosperms. Ann. Missouri Bot. Gard. sity Press, Cambridge. 441–450. [Presentation of an easy-to-follow 81: 419–450. Goloboff. 1993. NONA, version 1.5.1. explicit method to construct an evolution- *Eldredge, N. and J. Cracraft. 1980. Phylo- Distributed by the author, Tucuman, ary taxonomic classification.] genetic patterns and the evolutionary process: Argentina. *Backlund, A. and K. Bremer. 1998. To be or not Methods and theory in comparative biology. *Hennig, W. 1966. Phylogenetic systematics. to be: Principles of classification and mono- Columbia University Press, New York. University of Illinois Press, Urbana. typic plant families. Taxon 47: 391–400. *Farris, J. S. 1974. Formal definitions of para- Hibbett, D., and M. J. Donoghue. 1998. Bremer, K. 1983. Angiosperms and phyloge- phyly and polyphyly. Syst. Zool. 23: Integrating phylogenetic analysis and clas- netic systematics. Verh. Naturwiss. Ver. 548–554. [A polyphyletic group is defined sification in fungi. Mycologia 90: 347–356. Hamburg 26: 343–354. [Discussion of reticu- as “A group in which the most recent com- Hillis, D. M., M. W. Allard and M. M. Miya- late evolution and cladistics.] mon ancestor is assigned to some other moto. 1993. Analysis of DNA sequence Bremer, K. 1988. The limits of amino acid group and not the group itself.”] data: Phylogenetic inference. Methods sequence data in angiosperm phylogenetic Farris, J. S. 1979. The information content of Enzymol. 224: 456–487. reconstruction. Evolution 42: 795–803. the phylogenetic system. Syst. Zool. 28: *Huelsenbeck, J. P. 1995. Performance of phy- [Decay indices.] 458–519. logenetic methods in simulation. Syst. Biol. Bremer, K. and H.-E. Wanntorp. 1979. Hier- Farris, J. S. 1989. Hennig86. Version 1.5. Port 44: 17–48. [Comparison of performance in archy and reticulations in systematics. Jefferson Station, New York. computer simulations of parsimony, maxi- Syst. Zool. 28: 624–627. Felsenstein, J. 1978. Cases in which parsimony mum likelihood, and overall similarity *Brooks, D. R. and D. A. McLennan. 1991. and compatibility methods will be posi- methods of tree construction.] Phylogeny, ecology, and behavior. University tively misleading. Syst. Zool. 27: 401–410. *Huelsenbeck, J. P. and D. M. Hillis. 1993. of Chicago Press, Chicago. [Excellent pre- Felsenstein, J. 1981. Evolutionary trees from Success of phylogenetic methods in the sentation of biological uses of phylogenet- DNA sequences: A maximum likelihood four-taxon case. Syst. Biol. 42: 247–264. ic hypotheses.] approach. J. Mol. Evol. 17: 368–376. [Maximum likelihood and parsimony Cronquist, A. 1987. A botanical critique of Felsenstein, J. 1989. PHYLIP 3.2 manual. Univer- methods are both relatively successful in cladism. Bot. Rev. 53: 1–52. sity of California Herbarium, Berkeley. reconstructing phylogeny.] *Dahlgren, R., and F. N. Rasmussen. 1983. *Forey, P. L., C. J. Humphries, I. L. Kitching, R. Humphries, C. J. and V. A. Funk. 1982. Monocotyledon evolution: Characters and W. Scotland, D. J. Siebert and D. M. Cladistic methodology. In Current concepts phylogenetic estimate. In Evolutionary biol- Williams. 1992. Cladistics: A practical course in plant taxonomy, V. H. Heywood and D. ogy, vol. 16, M. K. Hecht, B. Wallace and G. in systematics. Oxford University Press, M. Moore (eds.), 323–362. Systematics T. Prance (eds.), 255–395. [Contains a sim- Oxford. [Summary of then-current cladis- Assn. Special Volume no. 25. Academic ple introduction to cladistic methods.] tic methods.] Press, London. [Introduction to basic Davis, P. D. and V. H. Heywood. 1973. Principles *Frohlich, M. W. 1987. Common-is-primitive: cladistic methods.] of angiosperm taxonomy. Krieger, New York. A partial validation by tree counting. Syst. Humphries, C. J. and L. R. Parenti. 1986. de Queiroz, A., M. J. Donoghue and J. Kim. Bot. 12: 217–237. [Given an ingroup and Cladistic biogeography. Clarendon Press, 1995. Separate versus combined analyses outgroup both with states a and b of a Oxford. of phylogenetic evidence. Annu. Rev. Ecol. homologous character, if state a is much Kellogg, E. A. 1989. Comments on genomic Syst. 26: 567–581. more common than state b in the outgroup, genera in the Triticeae. Am. J. Bot. 76: *de Queiroz, K. and J. Gauthier. 1990. Phylo- then state a is likely to be ancestral within 796–805 geny as a central principle in taxonomy: the ingroup.] Kellogg, E. A., R. Appels and R. J. Mason- Phylogenetic definitions of taxon names. Funk, V. A. 1985. Phylogenetic patterns and Gamer. 1996. When genes tell different sto- Syst. Zool. 39: 307–322. [Proposal to aban- hybridization. Ann. Missouri Bot. Gard. 72: ries: The diploid genera of Triticeae don the Linnaean system.] 681–715. (Gramineae). Syst. Bot. 21: 321–347. *Funk, V. A. and D. R. Brooks. 1990. Phylogenetic Kron, K. A. and W. S. Judd. 1997. Systematics *Items marked with an asterisk are especial- systematics as the basis of comparative biology. of the Lyonia group (Andromedeae, Smithsonian Contributions to Botany no. 73. Ericaceae) and the use of species as termi- ly recommended to those readers who are Washington, DC. [Examples of biological nals in higher-level cladistic analyses. Syst. interested in further information on the top- uses of cladograms.] Bot. 22: 479–492. ics discussed in Chapter 2. METHODS AND PRINCIPLES OF BIOLOGICAL SYSTEMATICS 33

*Maddison, W. P. and D. R. Maddison. 1992. Sneath, P. H. A. and R. R. Sokal. 1973. uted by Sinauer Associates, Sunderland, MacClade: Analysis of phylogeny and charac- Numerical taxonomy. W. H. Freeman, San MA.] ter evolution. Version 3.0. Sinauer Asso- Francisco. *Swofford, D. L., G. J. Olsen, P. J. Waddell and ciates, Sunderland, MA. [A useful program *Sokal, R. R. and F. J. Rohlf. 1980. An experi- D. M. Hillis. 1996. Phylogenetic inference. for exploring patterns of character change ment in taxonomic judgement. Syst. Bot. 5: In Molecular systematics, 2nd ed., D. M. on a cladogram.] 341–365. Hillis, C. Moritz and B. K. Mable (eds.), Maddison, W. P., M. J. Donoghue and D. R. Stevens, P. F. 1980. Evolutionary polarity of 407–514. Sinauer Associates, Sunderland, Maddison. 1984. Outgroup analysis and character states. Annu. Rev. Ecol. Syst. 11: MA. [Excellent summary of methods of parsimony. Syst. Zool. 33: 83–103. 333–358. tree construction.] Mayr, E. and P. D. Ashlock. 1991. Principles of *Stevens, P. F. 1984. Homology and phyloge- *Wagner, W. H., Jr. 1980. Origin and philoso- systematic zoology. 2nd ed. McGraw-Hill, ny: Morphology and systematics. Syst. Bot. phy of the groundplan–divergence New York. 9: 395–409. method of cladistics. Syst. Bot. 5: 173–193. *McDade, L. A. 1990. Hybrids and phyloge- Stevens, P. F. 1986. Evolutionary classification Wagner, W. H., Jr. 1983. Reticulistics: The netic systematics. I. Patterns of character in botany, 1960–1985. J. Arnold Arbor. 67: recognition of hybrids and their role in expression in hybrids and their implica- 313–339. cladistics and classification. In Advances in tions for cladistic analyses. Evolution 44: *Stevens, P. F. 1991. Character states, morpho- cladistics: Proceedings of the second meeting of 1685–1700. logical variation, and phylogenetic analy- the Willi Hennig Society, N. I. Platnick and *McDade, L. A. 1992. Hybrids and phyloge- sis: A review. Syst. Bot. 16: 553–583. V. A. Funk (eds.), 63–79. Columbia netic systematics. II. The impact of hybrids Stevens, P. F. 1998. What kind of classification University Press, New York. on cladistic analyses. Evolution 46: should the practising taxonomist use to be Wiley, E. O. 1979. An annotated Linnaean hier- 1329–1346. saved? In Plant diversity in Malesia III: archy, with comments on natural taxa and McDade, L. A. 1997. Hybrids and phylogenet- Proceedings of the 3rd International Flora competing systems. Syst. Zool. 28: 308–337. ic systematics. III. Comparison with dis- Malesiana Symposium 1995, J. Dransfield, *Wiley, E. O. 1981. Phylogenetics. John Wiley & tance methods. Syst. Bot. 22: 669–683. M. J. E. Coode and D. A. Simpson (eds.), Sons, New York. [Detailed discussion of Nelson, G. and N. Platnick. 1981. Systematics 295–319. Royal Botanical Gardens, Kew. cladistic principles.] and biogeography. Columbia University Stuessy, T. F. 1983. Phylogenetic trees in plant *Wiley, E. O., D. Siegel-Causey, D. R. Brooks Press, New York. systematics. Sida 10: 1–13. and V. A. Funk. 1991. The compleat cladist: A *Quicke, D. L. J. 1993. Principles and tech- Stuessy, T. F. 1990. Plant taxonomy. Columbia primer of phylogenetic procedures. University niques of contemporary taxonomy. Black- University Press, New York. of Kansas, Museum of Natural History, well Academic and Professional, London. *Swofford, D. L. 1993. PAUP: Phylogenetic Special Publ. no. 19. Lawrence, Kansas. *Sanderson, M. J. and M. J. Donoghue. 1989. analysis using parsimony. Version 3.1.1. [Summary of then-current cladistic meth- Patterns of variation in levels of homo- Distributed by the Illinois Natural History ods.] plasy. Evolution 43: 1781–1795. Survey, Champaign. [PAUP*: Phylogenetic Simpson, G. G. 1961. Principles of animal taxono- analysis using parsimony and other methods, my. Columbia University Press, New York. Version 4.0 is in beta-test edition, distrib-